Case
Hidden issue content can redirect Copilot into its available credentials
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
researchAGAINST AI3 claims
AI security research / cases
Case files
Investigations, evidence and outcomes.
369 Cases · Page 19 of 19
9 matching records on this page
Published cases
Case
JoySafeter client-selected Copilot mode reaches an unsandboxed host shell
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
CaseAGAINST AI1 claims
Case
Letta accepts a caller-selected tenant identity even with shared-secret authentication
Setting: research demonstrationExploitation: demonstratedEvidence: Grade B
CaseAGAINST AI1 claims
Case
Letta default self-hosting exposes the agent execution service without authentication
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
CaseAGAINST AI1 claims
Case
Letta local tools inherit the server credential environment
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
CaseAGAINST AI1 claims
Case
Multiple approval monitors can share one poisoned evidence boundary
Setting: evaluationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
researchAGAINST AI3 claims
Case
OpenSRE alert text can reach a shell tool that never requests approval
Setting: research demonstrationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
CaseAGAINST AI1 claims
Case
Persistent and shared memory extend injection across conversations and users
Setting: evaluationExploitation: demonstratedEvidence: Grade CRemediation: mitigation available
researchAGAINST AI3 claims
Case
Scanner selection rules can silently exclude vulnerable code before model review
Setting: evaluationExploitation: demonstratedEvidence: Grade BRemediation: mitigation available
researchAGAINST AI3 claims
No matches on this page.